Gas Turbines Are Holding Back the AI Revolution

“Buy one now and delivery is five years away, yet they still choose not to expand capacity.”

— Tang Yibo, on gas turbine manufacturers’ willingness to expand

Guest: Tang Yibo | Research analyst at a private investment fund and author of the WeChat publication “奔波儿r”

Orders for a gas turbine are already backed up five years, yet manufacturers’ expansion plans remain restrained. When data centers cannot buy large machines, they buy smaller ones. When those are insufficient, fuel cells, gas engines, and even retired aircraft engines converted into generators become alternatives.

My original understanding of power shortages was simple: install more solar and add storage. After Tang Yibo walked through the arithmetic, I realized that how much electricity is generated over a year and whether it is available at the precise moment you need it are very different questions. China has addressed such a gap before. The United States is trying a different approach, while Europe is developing structural problems of its own.

Tang began researching renewable energy in 2018. Following the bottlenecks in solar took him into grids, power plants, generating equipment, and electricity regulation. In this conversation, he explains how an industry shaped by a prolonged downturn suddenly collided with AI-driven data center demand: why three equipment makers hesitate to expand, why taking turbine blades from drawings to orders can require three or four years, and why a disused power plant can command a high price.

By the end, the discussion returns to stocks. The power-shortage story is big and easy to understand. But once investors have searched the entire supply chain, how much room does a good story still leave for making money?

Recorded on August 12, 2026. References to “this year,” “now,” and market conditions reflect the recording date.

The following is a full English translation of the Chinese conversation, which was edited and condensed for clarity.

1. Why Can Power Still Run Short as More Solar Is Installed?

Raymond: You were an energy analyst before this wave of AI. When did you start covering the industry?

Tang Yibo: In 2018. I had just graduated and joined Taikang Asset Management. My first assignment was solar. China had just introduced the May 31 subsidy cuts, and the whole sector had plunged. I started at the worst point in the cycle.

Raymond: How did you end up with that industry?

Tang Yibo: As a new graduate at a large institution, you do not get to choose. Your manager assigns you a sector. Someone already covers the attractive industries. Food and beverages, Moutai, and pharmaceuticals were doing exceptionally well then. Those were not going to be yours. I was assigned solar and environmental protection. I later stopped covering the latter.

Taikang placed a lot of emphasis on value investing. Research always asked what an industry would look like in five or ten years, what its end state would be, and how its business model worked. Looking back through solar’s history, I thought its business model was terrible. Two features stood out: “the leader is doomed” and “the latecomer’s advantage.”

Over the decade or more before 2018, successive changes in technology could send the former leader straight into bankruptcy. It might not even have a chance to retreat to second or third place. Even within the same technology, a later entrant could buy cheaper equipment and make more efficient products. You bought your equipment this year; next year, someone else could buy better equipment for less and manufacture the same products. For example, their equipment might cost a third less and have efficiency a fifth to a quarter higher. Output per unit of capital spending would differ enormously. Beyond changes in technology, early investors were continually being overtaken within the same technology. As a business model, that was hard for me to like.

Raymond: You later wrote about a global power shortage. How did your focus shift from solar to that?

Tang Yibo: Researching renewables in China leads you to the grid as a bottleneck. Solar and wind generate intermittently. The grid cannot accommodate unlimited amounts of them. Follow that problem and you notice the power rationing in 2021.

Many people’s instinct is: China has so much coal-fired generation and has installed so much renewable capacity, so why would power run short? Renewables were growing rapidly, but new coal-fired capacity had been constrained. Total installed generating capacity looked as though it was growing strongly, while dependable supply was not keeping pace.

The first thing to establish is: what, exactly, are we short of? In terms of annual electricity generation, solar and wind both add energy. But suppose it is five in the afternoon and you need electricity right now. It is getting dark and the wind is weak. Solar and wind cannot deliver what you need. That is when you need controllable sources such as nuclear, hydro, coal, or gas.

Electrical energy is measured in kilowatt-hours; load is measured in kilowatts. Load is power at a particular instant: however much electricity I need this second, you have to supply it.

Raymond: So even 1 gigawatt of coal-fired capacity should not be counted as 1 gigawatt available at any moment?

Tang Yibo: You have to allow for reserves and start-up or shutdown time. You cannot count 100%. As an illustration, think of 0.8 gigawatts as available. The contribution from wind and solar when they are needed is discounted much more heavily. You cannot simply add up their nameplate capacities.

You also need to consider the ratio of controllable supply to peak load. Under the framework I was using at the time, it should generally exceed 1.2, leaving a 20% margin. As that ratio falls, the power system becomes vulnerable to shortages.

Raymond: But as a residential customer, I barely noticed anything.

Tang Yibo: Households receive priority. Your air conditioner and lights may be unaffected, while a factory feels “staggered production” very directly: you operate during these hours, and another factory takes its turn afterward. Limited impact on households does not mean there is no shortfall for producers.

Raymond: Factories have to coordinate their operating hours. We added overall generation, but not enough of the supply available whenever it is needed.

Tang Yibo: Exactly. Economic development and people’s living needs keep pushing up the underlying load, by roughly five or six percent a year. That demand is steady. If dependable generation does not keep up, a gap emerges.

2. Does Putting a Battery Beside a Solar Farm Solve the Problem?

Raymond: How did China close that gap?

Tang Yibo: The response was very direct. I remember the phrase “three rounds of 80 GW”: approving 80 gigawatts of coal-fired capacity each year for several consecutive years. I do not remember the exact approval years, but the central point was to replenish dependable supply.

China builds quickly. Once approved, a plant can be completed in a little over a year. I think that demonstrates an advantage of the system. The transition toward renewables had previously been too fast and aggressive. Once the need for dependable supply was recognized, nuclear construction continued, gas-fired generation also developed, and coal-fired construction returned to a higher level.

Raymond: Could the underlying load still outgrow supply again?

Tang Yibo: I do not think China needs to worry too much about the same problem recurring. Coal-fired construction had been pushed too low. Even as annual additions decline from those 80-gigawatt rounds, they remain much higher than before. Continuing to add dependable supply should accommodate load growth.

Raymond: I have watched films about Chinese infrastructure that explain how storage compensates for unstable wind and solar. I took that to mean solar plus storage becomes a stable power source. Is that right?

Tang Yibo: The direction is right, but the scale matters. With storage, you have to look at both power and duration, rather than just how much has been installed.

For example, 1 gigawatt of solar might be paired with storage power equal to 20% to 40% of its capacity: 0.2 to 0.4 gigawatts, with two hours of duration. That mainly balances supply within a day. Solar produces more than you can use at noon; you store the excess and use it for another two hours in the evening.

But what if today is sunny and tomorrow is overcast? How much storage do you need to move today’s electricity into tomorrow, or across two days or a week? That is a completely different calculation.

One estimate I encountered in my research was that 1 gigawatt of solar paired with 1 gigawatt of storage power and ten hours of duration—10 gigawatt-hours of batteries—might produce a power source with a controllability coefficient of roughly 0.6 to 0.7. That would be very expensive.

Raymond: The cost explodes. It explodes.

Tang Yibo: The original calculation was a fraction of a gigawatt multiplied by two hours. Now it is one multiplied by ten. Chinese companies operating mines in Africa also build their own solar-plus-storage systems because the local grid is unreliable. In projects I have looked at, serving round-the-clock demand required roughly 16 gigawatt-hours of batteries per gigawatt of solar. Even then, supply reliability was only about 90%.

Raymond: I really had not done that calculation. I just assumed a battery next to a solar farm would solve it.

Tang Yibo: Balancing within a day and supplying power around the clock impose very different requirements.

Raymond: China filled its gap quite quickly with coal-fired generation. Why did the United States run into this problem too?

Tang Yibo: The starting conditions in the United States and Europe were different. US electricity demand had been flat for roughly a decade, and Europe’s for even longer. Without demand growth, there was no need to keep building dependable generating capacity. Install some solar, let gas plants run a little less, and the existing system could remain balanced.

After the pandemic, US manufacturing investment began to rise. Semiconductor, electric-vehicle, and battery factories received subsidies and started construction. AI was not the original driver. The United States had gone many years without adding new supply on a large scale. Once new demand arrived, the previously balanced system became tight. AI then pushed demand higher still.

Raymond: China’s dependable supply had not kept up; in the United States, demand suddenly increased.

Tang Yibo: Right. This US cycle was mainly demand-driven. When demand was not growing, a lot of the problems were hidden.

3. When Do Higher Electricity Prices Start Benefiting Equipment Makers?

Raymond: What is the American solution?

Tang Yibo: It is difficult to respond as China did, simply authorizing three rounds of 80 gigawatts and starting construction. China has workers and equipment available. In the United States, the response primarily works through the market: electricity prices rise first, encouraging more investment in generation.

The problem is that this process is slow. Rising demand does not immediately turn into gas turbine orders. First, the existing surplus is absorbed. Then electricity prices rise and power plant profits improve. Only once conditions reach a certain point do generators have an incentive to invest in new plants and equipment.

The United States uses a lot of natural gas for generation. The role of gas turbines there is comparable to the role of generating equipment in China’s coal-fired plants. Gas turbine orders did not clearly improve until the middle of 2024. Several years passed between the post-pandemic increase in manufacturing investment and the arrival of equipment orders.

Raymond: Power plant owners made money first?

Tang Yibo: Yes. Nuclear generator CEG and gas-fired generator VST had suffered during the downturn. Rising electricity prices improved their earnings, and their shares responded first. Only when they decided to build plants and buy equipment from GEV did the benefit reach equipment makers.

At first, buyers might also draw down inventories, so manufacturers did not feel the demand as strongly. Once inventories were gone and customers needed new machines manufactured, the shortfall became obvious.

Raymond: By the time someone asked GEV to make something new, it was even later.

Tang Yibo: Exactly. The previous downturn had hurt both generators and equipment makers badly. They were not going to expand the moment things improved a little. Even now, expansion is restrained.

Raymond: What kind of company is GEV? It is a newly listed entity spun out of General Electric, but the business itself is old.

Tang Yibo: GEV is GE Vernova. What we are mainly discussing here is its gas turbine business, which makes equipment for natural-gas-fired power generation. This industry is highly concentrated. General Electric, Siemens Energy, and Mitsubishi account for roughly 85% of the global market.

More importantly, those are the three names you see in current equipment sales, and they are also the names you see in cumulative installations since gas turbines became commercial. Over decades, almost no new entrant has changed the competitive structure.

Raymond: Chinese manufacturers did not enter?

Tang Yibo: One reason is resources. China has abundant coal but relatively little gas. Coal could meet much of its demand, so there was not the same strong domestic demand for gas-fired generation to drive the industry’s development.

The other reason is technology. Making gas turbines is comparable in difficulty to making aircraft engines. Some manufacturing capabilities overlap: companies that can make aircraft-engine blades can generally also make gas turbine blades, although dimensions and specific requirements differ. Aircraft engines need high thrust in a small package; gas turbines need continuous, reliable operation. Combustion temperatures are high, and the equipment must run for long periods, placing demanding requirements on materials, blades, and manufacturing processes.

That is why gas turbines and aircraft engines are called the crown jewels of industrial manufacturing. Resource endowments and technological barriers together produced today’s competitive structure.

Raymond: It sounds like the kind of business we usually want to find: high barriers, very few competitors, and familiar players who have been around for years.

Tang Yibo: The structure is stable and the barriers are high. But a good business model does not guarantee profits at every point in the cycle. You still need to look at supply and demand. In the last cycle, they lost money, laid people off, and closed factories too.

4. Why Could a 40% Price Cut Not Win the Maintenance Business?

Raymond: You wrote that GE could not win competitors’ customers even with a 40% price cut. Based on Chinese manufacturing experience, you would expect one of the three to start a price war. Why did that not work here?

Tang Yibo: First, to be clear, it was trying to win service market share. Gas turbines have a different business model from ordinary industrial equipment. Equipment and service are sold together, much as they are with aircraft engines.

A machine comes with a service contract lasting roughly 15 years. The equipment has a long life and operates in harsh conditions, requiring minor repairs, major overhauls, and replacement parts. A power plant finds it difficult to handle all that maintenance itself, so it generally signs a contract with the original manufacturer.

Raymond: Like AppleCare?

Tang Yibo: Much harder to do without than AppleCare. Specialized repairs and spare parts are involved. Add up the annual service payments and they can exceed twice the equipment’s sale price.

As an illustration, rather than a quote for a particular model, equipment might sell for $10 million and the subsequent service contract for $20 million to $25 million. Before the latest price increases, equipment could have a low or zero gross margin. The money mainly came from subsequent service, where the contract’s operating margin could be around 40%.

Raymond: So the lifetime value of one machine might be $35 million?

Tang Yibo: Yes: $10 million from equipment and another $20 million to $25 million from 15 years of service. That was the part of competitors’ businesses GE wanted to capture.

Raymond: But why is service so sticky? Why can’t I maintain it for you at a lower price?

Tang Yibo: The most profitable part of service is actually the spare parts. It is a little like getting a car repaired at an authorized dealership. Buying a component directly from the parts factory and having it replaced at the dealership are priced differently.

But these three companies’ machines are different. To maintain a Siemens machine, GE needs parts for Siemens models. Manufacturing a competitor’s components when you do not make the corresponding machines is uneconomic.

If you do not make the parts and instead buy them from the original manufacturer, that manufacturer has already taken the most profitable portion. What remains for you? Providing only labor and service is less attractive. Making the parts as well lacks sufficient scale to justify it.

I remember management setting targets for how many competitors’ machines they would service in the following quarter. They kept missing those targets, tried for a while, and eventually stopped pursuing it.

Raymond: Selling a machine with a 15-year maintenance contract sounds like great cash flow. Why did the industry still experience such a long downturn?

Tang Yibo: The sharp downturn began around 2015 or 2016, and conditions did not clearly improve until 2024. It was especially bad in 2018, when all three were making large layoffs and closing factories.

In any industry, supply and demand determine the state it is in, regardless of how damn good its business model or barriers to entry are. In the previous US gas-power construction cycle, demand was strong, so all three expanded. Once electricity demand stopped growing, demand for equipment fell off a cliff.

Raymond: Why does it drop that quickly? If annual demand used to be 100 gigawatts, wouldn’t it just fall to 80?

Tang Yibo: Equipment serves incremental demand. Suppose annual electricity consumption grows by 5% and needs 100 gigawatts of new capacity to meet it. If growth slows to 3%, perhaps you need only 60-something gigawatts. If growth falls to zero, you no longer need new generating capacity.

The plants already built are relatively new, so they do not yet need replacement either. Annual new-equipment demand can fall from 100 to 10 or 20 while manufacturing capacity remains. Even with only three suppliers, you get oversupply and price competition. Equipment may even be sold at a negative gross margin.

5. Why Hold Back When Orders Are Already Five Years Out?

Raymond: I once heard GEV management talk about expansion. The gist was: can anyone tell me which party will run the next government? That concern is straightforward.

Tang Yibo: Yes. The previous cycle hurt them too deeply. These businesses used to sit inside larger groups. Losses were severe, investors demanded divestments, and they eventually became separately listed companies.

US electricity demand was flat for years, and renewable-energy policies also weakened incentives to build new gas-fired plants. Manufacturers carried their existing capacity while demand contracted, eventually forcing layoffs and factory closures. Tell them today that another major cycle is beginning, and they are unlikely to believe it immediately.

Raymond: So the downturn mainly came from disappearing demand, not a new wave of competitors.

Tang Yibo: Right. Demand shrank rapidly, and supply followed it down. I remember all three announcing layoffs of roughly 30% and the closure of about a third of their factories in 2018. That was not so long ago.

Raymond: Even with this round of AI demand so visible, the expansion plans are still modest?

Tang Yibo: AI’s impact on electricity demand is enormous, and expectations keep rising each year. Existing gas turbine capacity cannot meet it, but the plans they have announced are roughly 25% growth over three years. If capacity starts at 50 gigawatts, it becomes 62.5 after three years. Very conservative.

Raymond: A Chinese manufacturer facing the same shortage would respond very differently.

Tang Yibo: I think management matters too. These are old companies that have undergone many changes and are run by professional managers. From a professional manager’s perspective, keeping the cycle strong may be more comfortable than aggressive expansion. Buy one now and delivery is five years away, yet they still choose not to expand capacity.

Raymond: Someone has to fill the gap.

Tang Yibo: That is how the whole “North American power-shortage supply chain” emerged. Buyers first look for large, heavy-duty gas turbines. If those are unavailable, they turn to smaller, light-duty turbines.

A heavy-duty unit might be around 500 megawatts, while a smaller unit might be 50. The latter were originally used more for on-site power at oilfields. If you cannot buy one big machine, buy ten small ones. Their generating costs are higher, but at least they can be delivered. Baker Hughes benefited from that demand.

At the approximate costs we were looking at then, heavy-duty turbines generated electricity for about $0.06 per kilowatt-hour, versus about $0.08 for smaller units. If those are insufficient too, buyers keep looking.

Raymond: What comes next?

Tang Yibo: Fuel cells from companies such as Bloom Energy. Their cost was at least $0.12 per kilowatt-hour at the time, considerably more than heavy-duty turbines. But lead times were short: three to six months. Installation was straightforward too. They could be stacked, rather like drawers.

Early projects were more often hospitals, schools, or small campuses, sometimes as supplemental power. At that point, supplying 500 megawatts or 1 gigawatt in a single deployment had not been validated at a comparable scale. But cloud companies could no longer wait.

Raymond: They were buying a more expensive, less fully proven solution because they could get it faster.

Tang Yibo: Yes. After that come gas engines from Caterpillar, Cummins, and Weichai. Some diesel-engine designs can be adapted to burn natural gas. They are smaller and less efficient, but they can generate electricity.

Then there is FTAI, which was in aircraft-engine-related businesses and converted engines for power generation. Previously, it was hard to imagine all these products being used to supply data centers. Now they have all appeared.

Prices rise, and strange sources of supply appear. Prices rise, and strange sources of supply appear. It is a very complete supply curve: the shortfall spreads from heavy-duty turbines into progressively more expensive, smaller equipment.

6. Does the Cycle Turn on Factory Expansion—or a Single Blade?

Raymond: Listening to this, if a private company has to solve its own power problem, it will do all sorts of crazy things to get electricity. How does the cycle end? Do these engines at the end of the chain stop selling first?

Tang Yibo: I think the real ending still depends on heavy-duty turbine expansion. If those machines can be delivered quickly, why choose the more expensive alternatives?

GEV is currently talking about 25% expansion over three years. If it could double capacity in that period, many of the alternatives would lose their economic rationale. One heavy-duty turbine replaces many smaller units. That determines when supply can increase substantially.

Raymond: So the key to watching this chain is when GEV, Siemens Energy, and Mitsubishi expand.

Tang Yibo: Yes, but even if they want to, supply-chain bottlenecks remain, particularly blades. If blade capacity cannot expand, the turbine manufacturers cannot expand either. Most of that supply chain is overseas, so you cannot look only at the final equipment makers’ intentions.

We began researching blades because Siemens Energy management said on an earnings call that blades were the bottleneck to expansion and that it was working with suppliers to resolve it.

The major suppliers in that field include Howmet and PCC, which make high-end castings. Production depends on skilled workers. More orders do not instantly create more capacity. When we researched Howmet, revenue had grown considerably, but headcount had not returned to its pre-pandemic level.

Raymond: The company is in the United States?

Tang Yibo: Yes. It laid people off during the pandemic, and hiring them back was not easy.

For Chinese suppliers, that extreme shortage created an opportunity. Under normal conditions, customers have little reason to risk changing suppliers. Gas turbines are expensive, and blades are critical components. A failure affects the operation of the whole power plant. Customers will not casually agree when you say, “Let’s try a Chinese supplier.”

Raymond: They become willing to try someone new only when the existing supplier genuinely cannot deliver.

Tang Yibo: Exactly. We looked at the domestic supply chain and found companies including Yingliu and Himile. Yingliu makes hot-section blades for gas turbines. They must operate at high temperatures, rotate continuously, and remain stable for long periods. The requirements are extremely demanding.

Yingliu already made castings and had customer relationships with GE and Caterpillar. It wanted to move into higher-end castings, so it approached GE. When it began working on blades in 2015, the customer gave it a drawing and told it to try. No money, and no large order to get it started.

It took roughly a year to make the first sample. One sample—not a batch of samples.

Raymond: Once that one blade was made, could it be installed in a turbine for testing?

Tang Yibo: One blade does not qualify for testing in a machine. It only proves that you can manufacture the thing. Yield and volume production are not even part of the discussion yet.

Some samples fail performance checks as soon as they are made, so there is no point discussing the next step. Once the first sample passes, you make a small batch to demonstrate that you can manufacture it with some consistency. A gas turbine needs more than one blade. How can you validate it with just one?

Raymond: I thought they would put that one blade into someone else’s machine. This is incredibly difficult.

7. Why Spend Ten Years on a Drawing That Comes With No Payment?

Tang Yibo: After that, there are small batches, tests, and validation in a machine, one step at a time. Initial yields are very low: perhaps one successful piece out of ten. You have to prove that you can produce in volume before the customer gives you further opportunities.

Once the parts are installed, they have to run for a long time to establish reliability. This equipment is meant to last for years. Working properly today does not mean a component has passed validation. The customer needs confidence that it will remain reliable over a long operating period.

From receiving the first drawing to securing an order that actually pays, the process takes roughly three to four years. Even then, the orders are only for small batches and smaller models. That is already relatively fast. To enter another customer’s system—Siemens, for example—you go through validation again.

Raymond: So there is no shortcut where you say, “I’ve already made it for GE,” and everyone else accepts it.

Tang Yibo: Correct. Different customers each have to validate it. You can see why catching up is so hard: someone else started years ago. If you notice the industry is attractive today and start from your first drawing, you still have a long road ahead.

Raymond: Where is Yingliu now?

Tang Yibo: From when we began following it, the stock’s gain reached fivefold at the peak and is now roughly threefold. It is already widely recognized. People know it is one of China’s leading gas turbine blade manufacturers. That lead did not appear overnight.

It had to satisfy several demanding conditions. First, it had to start very early. But second, at that time there was no domestic demand and overseas markets were in a downturn. You needed exceptional foresight or very strong conviction to keep investing.

When we looked at it, annual net profit was around RMB100 million, but annual R&D spending was RMB200 million to RMB300 million, and capital spending was another RMB200 million to RMB300 million. Its R&D expense ratio was high for a manufacturer, in the teens. It sustained that investment for ten years.

Raymond: That is a venture-capital approach to running a company.

Tang Yibo: Yes. We also visited the company. The chairman was already older, but whenever he discussed the technology, his eyes lit up. He became energetic and excited. Meeting him helped explain why he had been willing to make such a big commitment in 2015 and persist with something so difficult.

Raymond: And it was a deeply unfavorable cycle at the time.

Tang Yibo: Extremely unfavorable. There was no domestic demand, so it had to serve overseas customers; overseas markets were in a downturn. The difficulty was not just technological. It was whether you could keep doing it when demand was absent.

Raymond: All the companies we have mentioned have already risen a lot. Looking at the power-shortage chain today, is anything still early or not fully recognized? None of these names are investment recommendations, of course.

Tang Yibo: If we are talking only about equipment, I think it is already difficult. The logic has spread from heavy-duty turbines at GEV and Siemens Energy to smaller turbines, fuel cells, Caterpillar’s gas engines, and China’s Weichai.

Could you have imagined a year ago that Chinese engine makers would supply power for American data centers? By this stage, the supply-demand gap has essentially been addressed. Unlike heavy-duty turbines, engines do not face such a clear blade bottleneck, and their manufacturers are expanding aggressively. Each unit is much smaller in power output, but they can make up for that with numbers.

Raymond: I picture lots of little vehicles parked beside a data center, like ants feeding electricity into it.

Tang Yibo: Not vehicles—engines. They are installed in buildings and generate power unit by unit. Because they are small, you use many of them.

Raymond: Lots of units feeding this enormous creature.

Tang Yibo: Exactly. But looking at the stocks, equipment makers’ prices have already reflected a great deal of this.

8. Where Could Data Centers Go if They Cannot Be Built in America?

Raymond: What are you looking at beyond equipment?

Tang Yibo: One potentially earlier-stage direction is this: if data center construction becomes difficult in the United States, could some of it move elsewhere?

For building in the United States, equipment appears no longer to be in short supply. But local opposition is clearly visible. Data centers increase power demand, and residents ask what benefit they receive and why their electricity bills are rising. States have different policies. The federal government wants construction to move faster, but local governments have their own voters, policies, and opposition.

If that resistance delays construction, America may not build as many data centers as originally planned. Some of the demand would have to find somewhere else. Inference may need to remain close to users in the United States; training could move to more welcoming locations.

Raymond: So you are looking at generation, rather than equipment.

Tang Yibo: Yes. A new power plant in the Nordics might still use equipment from the same few manufacturers. But if data center demand cannot stop, moving to a place with more available electricity is one solution.

You can see plans for data centers and power plants in Japan, South Korea, and Southeast Asia. Europe is another possibility, although conditions vary greatly within it. Several areas where data centers have traditionally clustered already have tight grid capacity. In some places, no more can connect, so construction is shifting toward Southern Europe and the Nordics.

Southern Europe has solar; the Nordics have hydro and nuclear power. Places with more available electricity and grid capacity may attract new demand.

Raymond: What kind of company does that lead you to?

Tang Yibo: One Nordic generator we researched, for example, had a dividend yield of about 5% at the time and sold power on market terms. If demand and power prices rise, that can flow directly into its profits.

You need to distinguish two types of company. Market-based generators have earnings exposed to electricity prices. Regulated utilities earn mainly according to an allowed return.

Take a simple example: invest RMB10 billion and receive an allowed return of 8%, or RMB800 million. Changes in market power prices do not produce the same earnings sensitivity as they do for a market-based generator. Growth primarily comes from investing more and expanding the asset base.

Raymond: A plant that used to supply nearby residents suddenly gets a data center next door willing to pay more for electricity. That could change things for a market-based generator.

Tang Yibo: That is roughly the logic. But for it to show up in earnings, the data center has to be completed and consuming electricity. An investment announcement is not enough. There is a time lag.

Raymond: So relocation may already be visible in the real economy without being fully reflected in share prices.

Tang Yibo: That is one possibility. These companies used to be seen as defensive utility stocks. If demand really changes, the market may reassess them.

But do not assume moving to another country means you can connect to the grid wherever you like. In Japan, South Korea, and Europe, large cloud providers have already taken much of the available space. Apply for a grid connection today and in some places the wait is eight to ten years.

Raymond: Similar things happen in America. Cloud companies apply everywhere and build wherever a slot opens first.

Tang Yibo: Exactly. They still prefer a grid connection. Self-generation is a last resort. The grid has many sources backing one another up; if your own power source fails, you have a problem. So everyone is competing for a place to connect.

9. Can Twenty Years of Covering Power Make You Miss the Change?

Raymond: Regulated companies are less sensitive to electricity prices. Why do overseas analysts still like recommending them?

Tang Yibo: Because the earnings are more predictable. European electricity demand went many years without growth. That is the environment analysts are accustomed to. Regulated companies lack price sensitivity, but grids and dependable supply now genuinely need expansion. Invest more, and profits can grow under the regulatory framework.

Public infrastructure companies also have five-year plans. Knowing future investment and asset growth lets you estimate earnings per share. Add dividends, and you get something that looks stable with predictable growth. For instance, a company might previously trade at 15 times earnings and distribute most profits as dividends. New capital spending now lets analysts calculate growth corresponding to asset expansion. It may require borrowing or raising capital, but the growth path is planned and easier to assess than future market electricity prices.

Raymond: They are still selecting the most predictable earnings in a traditional sector.

Tang Yibo: Yes, rather than seeking the greatest upside sensitivity. I think there is also anxiety behind it: will the data centers actually get built? Will demand be unstable once they are completed? After watching a twenty-year downturn, it is hard to suddenly believe power prices can rise for a sustained period. Apart from the surge following the Russia–Ukraine conflict, they have not seen it for a very long time.

Raymond: Can knowing a sector for so long become a burden?

Tang Yibo: It happens often. Studying an industry continuously can make you miss an opportunity when it changes dramatically. Most traditional auto analysts I knew missed BYD and Tesla early on.

With BYD, they remembered its old product, dealer, and inventory problems. When the company really began to take off, they still did not believe it. The thought in their heads was: I have covered this industry for more than a decade; how could you know more than I do? I have never seen such a change in all that time. How could it happen?

But something not happening for more than a decade does not mean it cannot happen next year.

Raymond: You have to remind yourself not to become an out-of-touch old hand. Has Europe actually reached the point of a power shortage today?

Tang Yibo: Europe as a whole is behind the United States in this cycle. Years of deindustrialization kept demand weak. It has only recently begun recovering, without the clear overall supply gap seen in America. Structural problems, however, are already obvious.

At noon, abundant solar can make electricity very cheap. Later in the day, or in certain seasons, prices rise sharply. That is also why storage is doing well in Europe: it can exploit intraday price differences. An adequate overall supply does not mean every location and every hour are problem-free.

If more data centers are built, demand could rise further. The more conservative forecasts I have seen put electricity demand growth at around 2% annually over the next few years. Under more optimistic data center assumptions, it could exceed 3%. For a power system, that is substantial growth.

Raymond: America has already demonstrated this: a system balanced for years suddenly faces a few additional percentage points of demand, and new supply cannot keep up.

Tang Yibo: Right. If Europe experiences the same thing, I think changing policy and building generating capacity may take even longer than in the United States, potentially creating another gap.

Raymond: What is the earliest indicator to watch?

Tang Yibo: Grid connection applications. From what I remember, European data centers were applying for over 200 gigawatts at the start of 2025, over 300 by October, and roughly 450 by June or July 2026. The increase was rapid.

That is only the statistical measure I saw, and these are very preliminary applications. Some may be little more than inquiries. What they first show is growing demand to find electricity and secure a place.

10. What Does €400 Million Actually Buy?

Raymond: Do those applications mainly come from American companies?

Tang Yibo: The dataset I saw did not break down applicants. It aggregated grid data, so you cannot infer that directly from it.

America also saw a surge in applications, but you have to be careful about the definitions. On the generation side, a solar project queues to feed power into the grid. On the consumption side, a data center queues to draw power from it. Those are different things. You cannot count every application as data center load.

Some applications simply reserve a place in line. It is like joining Beijing’s queue for a new-energy vehicle license plate: whether you buy the car or not, secure the number first. There can therefore be duplicate applications or applications that never result in construction.

Raymond: Grid applications are an early indicator. Then you have to track actual construction and delivery.

Tang Yibo: Yes. You already see plenty of announcements about investment in Nordic and Southern European data centers, but ultimately you need to follow how much gets built. There are also some very specific behaviors: buying disused factories and retired power plants.

The factory may be dilapidated, but it has a transformer nearby and a grid connection. A buyer may value that connection rather than the building. Old assets left behind by Europe’s deindustrialization can become valuable again.

I saw one retired RWE thermal power plant that could no longer generate electricity, but had a grid connection and sold for €400 million.

Raymond: Wait. That €400 million bought the right to connect to the grid?

Tang Yibo: You can think of it that way. The land may not be worth that much, and the old generating equipment still needs dismantling. The key is the connection: buy it and you can use electricity. A new application could mean waiting eight to ten years.

Raymond: Effectively, you are discounting ten years of time value back to today.

Tang Yibo: Yes. Time matters. Data centers want to come online quickly, so of course they look for ways to reduce the wait.

Raymond: I like this style of research. It looks as though you are talking about macroeconomics and history, but you end up at specific companies. Yet everyone now seems to be talking about turbines, overseas expansion, and power plants. If you scored how crowded the trade is from one to ten, what would you give it?

Tang Yibo: Eight or nine. It is already crowded—very crowded.

This chain has been playing out since 2023, for three years now. Investors have worked through overseas companies and then domestic ones, mature and reliable generating technologies and then more marginal alternatives. By March and April this year, I kept seeing videos about “gas turbines going overseas.” That was when I felt I no longer wanted to keep trading stocks in this theme.

Raymond: I keep seeing it too. I do not even know the industry in great depth, and it already looks busy to me.

Tang Yibo: Its popularity feels second only to memory chips. The narrative is enormous: AI, electricity, energy, and supply bottlenecks. But from a stock-market perspective, I think it has already played out very fully. It is very hot and very crowded. It is hard to make money now.

If you're working on this too — or you think we've got it wrong — write to us at [email protected]; if you'd rather not write, just leave your email below.